Research Vision
Our long-term goal is to bridge modern AI and statistical science to build computational tools that are not only powerful and scalable but also trustworthy and interpretable. By combining the flexibility of AI with the rigor of statistics, we aim to drive transformative advances in biomedical research, enabling discoveries that were previously out of reach.
Research Overview
Our current research interests aim at understanding how genetic variation shapes gene regulation, cellular states, and ultimately human disease. A major challenge is that these processes are highly context dependent, varying across cell types, tissues, individuals, space, and time, and many of the molecular states connecting genotype to phenotype are only partially observed. We develop AI-powered statistical and computational methods for causal regulatory genomics to uncover these hidden regulatory mechanisms.
Our work spans several complementary directions. Context-aware genomic foundation models, including RegFM (Gao et al., bioRxiv 2026) and EpiGePT (Gao et al., Genome Biol. 2024), learn transferable regulatory knowledge from large-scale genomic data. Generative AI for single-cell and multiomics, including scDiffusion-X (Luo et al., Nat. Commun. 2026), scMTG (Cui et al., bioRxiv 2026), scDEC (Liu et al., Nat. Mach. Intell. 2021), and MultiFlow (Wang et al., bioRxiv 2026), reconstructs and predicts cellular states across modalities, biological contexts, time, and perturbations. Trustworthy causal AI, including CausalBGM (Liu et al., J. Am. Stat. Assoc. 2026), BGM (Liu et al., arXiv 2026), CausalEGM (Liu et al., PNAS 2024), and Roundtrip (Liu et al., PNAS 2021), aims to move beyond association/prediction toward causal effect estimation and mechanistic interpretation.
Together, these directions form an integrated research paradiagm for connecting genetic variation → regulatory effects → gene programs → cellular responses → phenotype, with the broader goal of turning increasingly large and complex genomic datasets into interpretable knowledge about human biology and disease.